Sei sulla pagina 1di 11

Chapter 04.

10
Eigenvalues and Eigenvectors

1. define eigenvalues and eigenvectors of a square matrix,

2. find eigenvalues and eigenvectors of a square matrix,
3. relate eigenvalues to the singularity of a square matrix, and
4. use the power method to numerically find the largest eigenvalue in magnitude of a
square matrix and the corresponding eigenvector.

What does eigenvalue mean?

The word eigenvalue comes from the German word Eigenwert where Eigen means
characteristic and Wert means value. However, what the word means is not on your mind!
You want to know why I need to learn about eigenvalues and eigenvectors. Once I give you
an example of an application of eigenvalues and eigenvectors, you will want to know how to
find these eigenvalues and eigenvectors.

Can you give me a physical example application of eigenvalues and eigenvectors?

Look at the spring-mass system as shown in the picture below.
k k
m1 m2

x1 x2

Assume each of the two mass-displacements to be denoted by x1 and x 2 , and let us assume
each spring has the same spring constant k . Then by applying Newton’s 2 nd and 3rd law of
motion to develop a force-balance for each mass we have
d 2x
m1 21  kx1  k ( x2  x1 )
dt

04.10.1
04.10.2 Chapter 04.10

d 2 x2
m2  k ( x2  x1 )
dt 2
Rewriting the equations, we have
d 2x
m1 2 1  k ( 2 x1  x2 )  0
dt
d 2 x2
m2 2  k ( x1  x2 )  0
dt
Let m1  10, m2  20, k  15
d 2x
10 21  15(2 x1  x2 )  0
dt
d 2 x2
20  15( x1  x2 )  0
dt 2
From vibration theory, the solutions can be of the form
xi  Ai sin  t  0 
where
Ai = amplitude of the vibration of mass i ,
 = frequency of vibration,
 = phase shift.
0
then
d 2 xi
2
  Ai w 2 Sin(t  0 )
dt
d 2 xi
Substituting xi and in equations,
dt 2
 10 A1 2  15(2 A1  A2 )  0
 20 A2 2  15( A1  A2 )  0
gives
(10 2  30) A1  15 A2  0
 15 A1  (20 2  15) A2  0
or
( 2  3) A1  1.5 A2  0
 0.75 A1  ( 2  0.75) A2  0
In matrix form, these equations can be rewritten as
  2  3  1.5   A1  0
     
  0.75    0.75  A2  0
2

 3  1.5  A1  2  A1  0 
 0.75 0.75   A     A   0
  2   2  
Let   
2

 3  1.5
[ A]  
 0.75 0.75 
Eigenvalues and Eigenvectors 04.10.3

A 
[X ]   1 
 A2 
[ A][ X ]  [ X ]  0
[ A][ X ]  [ X ]
In the above equation,  is the eigenvalue and [ X ] is the eigenvector corresponding to  .
As you can see, if we know  for the above example we can calculate the natural frequency
of the vibration
 
Why are the natural frequencies of vibration important? Because you do not want to have a
forcing force on the spring-mass system close to this frequency as it would make the
amplitude Ai very large and make the system unstable.

What is the general definition of eigenvalues and eigenvectors of a square matrix?

If [ A] is a n  n matrix, then [ X ]  0 is an eigenvector of [ A] if
[ A][ X ]  [ X ]
where  is a scalar and [ X ]  0 . The scalar  is called the eigenvalue of [ A] and [ X ] is
called the eigenvector corresponding to the eigenvalue  .

How do I find eigenvalues of a square matrix?

To find the eigenvalues of a n  n matrix [ A] , we have
[ A][ X ]  [ X ]
[ A][ X ]  [ X ]  0
[ A][ X ]   [ I ][ X ]  0
([ A]  [ ][ I ])[ X ]  0
Now for the above set of equations to have a nonzero solution,
det([ A]  [ I ])  0
This left hand side can be expanded to give a polynomial in  and solving the above
equation would give us values of the eigenvalues. The above equation is called the
characteristic equation of [ A] .
For a [ A] n  n matrix, the characteristic polynomial of A is of degree n as follows
det([ A]  [ I ])  0
giving
n  c1n1  c 2  n 2     c n  0
Hence. this polynomial has n roots.

Example 1
Find the eigenvalues of the physical problem discussed in the beginning of this chapter, that
is, find the eigenvalues of the matrix
 3  1.5
[ A]  
 0.75 0.75 
Solution
 3  1 .5 
[ A]  [ I ]  
 0.75 0.75   
04.10.4 Chapter 04.10

det( A    I  )  (3   )(0.75   )  ( 0.75)(1.5)  0

2.25  0.75  3  2  1.125  0
2  3.75  1.125  0
 ( 3.75)  ( 3.75) 2  4(1)(1.125)

2(1)
3.75  3.092

2
 3.421, 0.3288
So the eigenvalues are 3.421 and 0.3288.

Example 2
Find the eigenvectors of
 3  1.5
A
 0.75 0.75 
Solution
The eigenvalues have already been found in Example 1 as
1  3.421,  2  0.3288
Let
x 
[X ]   1 
 x2 
be the eigenvector corresponding to
1  3.421
Hence
([ A]  1 [ I ])[ X ]  0
 3  1.5 1 0  x1 
   3.421     0
 0.75 0.75   0 1    x2 
 0.421  1.5   x1  0
  0.75  2.671  x   0
  2   
If
x1  s
then
 0.421s  1.5 x 2  0
x 2  0.2808s
The eigenvector corresponding to 1  3.421 then is
 s 
[X ]   
 0.2808s 
 1 
 s 
  0 .2808
The eigenvector corresponding to
1  3.421
is
Eigenvalues and Eigenvectors 04.10.5

 1 
 0.2808
 
Similarly, the eigenvector corresponding to
 2  0.3288
is
 1 
1.781
 

Example 3
Find the eigenvalues and eigenvectors of
 1.5 0 1 
[ A]  
  0.5 0.5  0.5

  0.5 0 0 
Solution
The characteristic equation is given by
det([ A]  [ I ])  0
1.5   0 1 

det   0.5 0.5    0.5  0
  0.5 0   
(1.5   )[(0.5   )( )  (0.5)(0)]  (1)[(0.5)(0)  ( 0.5)(0.5   )]  0
 3  22  1.25  0.25  0
To find the roots of the characteristic polynomial equation
 3  22  1.25  0.25  0
we find that the first root by observation is
 1
as substitution of   1 gives
( 1) 3  2(1) 2  1.25(1)  0.25  0

00
So
(  1)

is a factor of
 3  22  1.25  0.25 .
To find the other factors of the characteristic polynomial, we first conduct long division
 2    0.25
  1    2  1.25  0.25
3 2

 3  2
______________________
04.10.6 Chapter 04.10

2  1.25  0.25
2  
 0.25  0.25
 0.25  0.25
Hence
 3  22  1.25  0.25  (  1)(2    0.25)

To find zeroes of  2    0.25 , we solve the quadratic equation,

 2    0.25  0
to give
 (1)  (1) 2  ( 4)( 1)(0.25)

2( 1)
1 0

2
 0.5,0.5
So
  0.5 and   0.5 are the zeroes of
 2    0.5
giving
 2    0.25  (  0.5)(  0.5)

Hence
 3  22  1.25  0.25  0
can be rewritten as
 (  1)(  0.5)(  0.5)  0

to give the roots as

  1, 0.5, 0.5

These are the three roots of the characteristic polynomial equation and hence the eigenvalues
of matrix [A].
Note that there are eigenvalues that are repeated. Since there are only two distinct
eigenvalues, there are only two eigenspaces. But, corresponding to   0.5 there should be
two eigenvectors that form a basis for the eigenspace corresponding to   0.5 .
Given
[( A  I )][ X ]  0
then
Eigenvalues and Eigenvectors 04.10.7

1.5   0 1   x1  0
  0 .5 0 .5    0.5  x 2   0

  0.5 0     x3  0
For   0.5 ,
 1 0 1   x1  0
 0.5 0  0.5  x   0
  2   
 0.5 0  0.5  x3  0
Solving this system gives
x1   a, x2  b, x3  a
So
 x1   a 
x    b 
 2  
 x3   a 
  a  0 
  0   b 
 a  0
 1 0 
 a  0   b 1 
 
 1  0
  1 0
  1 
So the vectors  0  and   form a basis for the eigenspace for the eigenvalue   0.5
1 
  0
 
and are the two eigenvectors corresponding to   0.5 .
For   1 ,
 0.5 0 1   x1  0
 0.5  0.5  0.5  x   0
  2   
 0.5 0  1   x3  0
Solving this system gives
x1  a, x 2  0.5a, x3  0.5a
The eigenvector corresponding to   1 is
 a   1 
  0.5a   a   0.5
   
  0.5a    0.5
Hence the vector
 1 
  0.5
 

  0.5

is a basis for the eigenspace for the eigenvalue of   1 , and is the eigenvector corresponding
to   1 .
04.10.8 Chapter 04.10

What are some of the theorems of eigenvalues and eigenvectors?

Theorem 1: If [ A] is a n  n triangular matrix – upper triangular, lower triangular or
diagonal, the eigenvalues of [ A] are the diagonal entries of [ A] .
Theorem 2:   0 is an eigenvalue of [ A] if [ A] is a singular (noninvertible) matrix.
Theorem 3: [ A] and [ A] T have the same eigenvalues.
Theorem 4: Eigenvalues of a symmetric matrix are real.
Theorem 5: Eigenvectors of a symmetric matrix are orthogonal, but only for distinct
eigenvalues.
Theorem 6: det( A) is the product of the absolute values of the eigenvalues of [ A] .

Example 4
What are the eigenvalues of
6 0 0 0 
7 3 0 0 
[ A]  
9 5 7.5 0 
 
2 6 0  7.2
Solution
Since the matrix [ A] is a lower triangular matrix, the eigenvalues of [ A] are the diagonal
elements of [ A] . The eigenvalues are
1  6,  2  3, 3  7.5, 4  7.2

Example 5
One of the eigenvalues of
5 6 2 
[ A]  3 5 9 
2 1  7 
is zero. Is [ A] invertible?
Solution
  0 is an eigenvalue of [ A] , that implies [ A] is singular and is not invertible.

Example 6
Given the eigenvalues of
 2  3.5 6 
[ A]  
3.5 5 2 

 8 1 
8.5
are
1  1.547,  2  12.33, 3  4.711
What are the eigenvalues of [B ] if
 2 3.5 8 
[B ]  
  3. 5 5 1 

 6 2 8.5

Eigenvalues and Eigenvectors 04.10.9

Solution
Since [ B ]  [ A]T , the eigenvalues of [ A] and [B ] are the same. Hence eigenvalues of
[B ] also are
1  1.547,  2  12.33, 3  4.711

Example 7
Given the eigenvalues of
 2  3.5 6 
[ A]  
3.5 5 2 

 8 1 8.5

are
1  1.547,  2  12.33, 3  4.711
Calculate the magnitude of the determinant of the matrix.
Solution
Since
det[ A]  1  2 3
  1.547 12.33 4.711
 89.88

How does one find eigenvalues and eigenvectors numerically?

One of the most common methods used for finding eigenvalues and eigenvectors is the
power method. It is used to find the largest eigenvalue in an absolute sense. Note that if this
largest eigenvalues is repeated, this method will not work. Also this eigenvalue needs to be
distinct. The method is as follows:
1. Assume a guess [ X ( 0 ) ] for the eigenvector in
[ A][ X ]  [ X ]
equation. One of the entries of [ X ( 0 ) ] needs to be unity.
2. Find
[Y (1) ]  [ A][ X ( 0 ) ]
3. Scale [Y (1) ] so that the chosen unity component remains unity.
[Y (1) ]  (1) [ X (1) ]
4. Repeat steps (2) and (3) with
[ X ]  [ X (1) ] to get [ X ( 2 ) ] .
5. Repeat the steps 2 and 3 until the value of the eigenvalue converges.
E
If s is the pre-specified percentage relative error tolerance to which you would like the
answer to converge to, keep iterating until
( i 1)  ( i )
 100  E s
( i 1)
where the left hand side of the above inequality is the definition of absolute percentage
relative approximate error, denoted generally by E s A pre-specified percentage relative
tolerance of 0.5  10 2  m implies at least m significant digits are current in your answer.
04.10.10 Chapter 04.10

When the system converges, the value of  is the largest (in absolute value) eigenvalue of
[ A] .

Example 8
Using the power method, find the largest eigenvalue and the corresponding eigenvector of
 1.5 0 1 
[ A]  
  0.5 0.5  0.5

  0.5 0 0 
Solution
Assume
1
[X ( 0)
]  1
1
 1.5 0 1  1
[ A][ X ( 0)
]
  0.5 0.5  0.5  
 1

  0.5 0 0 1

 2.5 
  0.5

 0.5 
 1 
Y (1)  2.5  0.2
  0.2
(1)  2.5
We will choose the first element of [ X ( 0 ) ] to be unity.
 1 
[ X ]   0.2
(1)

 0.2
 1.5 0 1  1 
[ A][ X ]    0.5
(1)
0 .5  0.5  0.2
  0.5 0 0   0.2
 1.3 
  0.5

 0.5 
 1 
[ X ( 2 ) ]  1.3 0.3846
 0.3846
( 2)  1.3
 1 
[X ( 2)
]    0.3846

  0.3846
The absolute relative approximate error in the eigenvalues is
Eigenvalues and Eigenvectors 04.10.11

( 2 )  (1)
a   100
( 2 )
1.3  1.5
  100
1.5
 92.307%
Conducting further iterations, the values of (i ) and the corresponding eigenvectors is given
in the table below

i (i ) [ X (i ) ]  a (%)
 1 
  0.2
1 2.5   _____
  0.2
 
 1 
  0.38462
2 1.3   92.307
  0.38462
 1 
  0.44827 
3 1.1154   16.552
  0.44827 
 1 
  0.47541
4 1.0517   6.0529
  0.47541
 1 
  0.48800
5 1.02459   1.2441
  0.48800
The exact value of the eigenvalue is   1
and the corresponding eigenvector is
 1 
[ X ]   0.5
 0.5

Key Terms:
Eigenvalue
Eigenvectors
Power method